Red Door Analytics KM reconstruct & extrapolate

Statistical tools

Reconstruct IPD from a published Kaplan–Meier curve — then extrapolate

Trace a published survival figure, recover the patient-level data (Guyot reconstruction), fit the standard parametric and flexible-parametric models, and extrapolate to a lifetime horizon. The lesson: curves that agree inside the trial fan out once the data run out — and the lifetime mean is governed by the model, not the data.

Load the worked example (a real breast-cancer trial) or trace a curve, then reconstruct to fit and extrapolate. The candidate models, AIC/BIC ranking and lifetime estimates appear here.

Red Door Analytics

Survival extrapolation for HTA

Choosing and justifying an extrapolation model — parametric, flexible-parametric, spline, or with external evidence and general-population mortality — is a core part of a cost-effectiveness submission. It is what merlin and our HTA work are for.

Talk to us

See also: Survival DGM explorer· Multi-state explorer· When the hazard ratio misleads· how we pre-specify an analysis →

Reconstruction follows Guyot et al. (2012) and is checked against an independent implementation of the method. Fits are by in-browser maximum likelihood and are checked against merlin on the German Breast Cancer Study data used in the worked example: the log-likelihood agrees to within 0.001 and the fitted survival, extrapolation included, to within 0.0015. Reconstructed data are pseudo-IPD — for teaching and exploratory analysis; for a submission, fit the models in merlin.